[Ns-developers] NS3 - Queue Scheduler

[Xiaoliang "David" Wei]

Hi Hagen,

On Tue, Apr 1, 2008 at 10:11 AM, Hagen Paul Pfeifer <hagen at jauu.net> wrote:
> * Mathieu Lacage | 2008-04-01 08:44:32 [-0700]:
>
>
>  >I am not really sure I understand your question exactly but I will try
>  >to answer anyway.
>  >
>  >1) In a parallel simulation where you run a set of separate parallel
>  >partitions, there is no constraint on which algorithm and data structure
>  >you use to sort and schedule events within that partition. So, working
>  >on parallel simulations does not mean adding new event scheduling
>  >algorithms and data structures
>
>  Right, no doubt. The only reason I asked was that there are some major changes
>  on the scheduler in ns-2 (2.33). If the answer was yes then I had studied had
>  to study the new algorithm too. So consider this question as an additional
>  hint. And of course Mathieu: I will not add any new event scheduling algorithm
>  or data structure - of course! ;)
>
>  Sorry for my misleading expressions - it was a trivial question about
>  changes in the scheduler algorithm in the near future.

I think the changes in 2.33 is more like an enhancement to the
Calendar Queue. The concepts are the same and the changes help the
algorithm to better estimate the event departure rate and the bin
size. (There is longer details documented at
http://netlab.caltech.edu/projects/ns2tcplinux/ns2patch/)

I also agree to Mathieu that the changes are orthogonal to parallel simulation.

-David

>
>
>  >2) Adding support for parallel simulations which is really transparent
>  >to the user would require that you can ensure the _exact_ same
>  >whole-simulation event ordering in both the parallel and non-parallel
>  >cases. To do this in a not-too-ineficient way will most likely require
>  >that you change the sorting key even in the non-parallel case so, that
>  >will require a few (hopefully small) changes to the current in-tree
>  >schedulers.
>
>  Right, so far I read ~5 papers of Fujimoto - so you can assume that I am more
>  or less familiar what must be done to parallelize an algorithm - at least
>  theoretical ... ;)
>
>  Best regards, Hagen
>
>
>  --
>  Hagen Paul Pfeifer <hagen at jauu.net>  ||  http://jauu.net/
>  Key Id: 0x98350C22
>  Key Fingerprint: 490F 557B 6C48 6D7E 5706 2EA2 4A22 8D45 9835 0C22
>  Always in motion, the future is.
>
>
>



-- 
Xiaoliang "David" Wei
http://davidwei.org
***********************************************